A discrimination analysis for unsupervised feature selection via optic diffraction principle

Area of Science:

• Machine Learning
• Data Analysis
• Signal Processing

Background:

• Feature selection is crucial for effective data analysis.
• Existing methods may not handle data orientation or scaling invariance well.

Purpose of the Study:

• To propose an unsupervised discrimination analysis for feature selection.
• To develop a method invariant under feature scaling.
• To extend the approach for handling data orientation.

Main Methods:

• Utilizes properties of the Fourier transform of probability density distributions.
• Employs an evaluation inspired by optical diffraction.
• Calculates a discrimination score for feature evaluation.
• Extends the method to account for data alignment and orientation.

Main Results:

• The proposed method demonstrates effectiveness on real-world datasets.
• Feature evaluation is achieved through a discrimination score.
• The approach is invariant to feature scaling.

Conclusions:

• The unsupervised discrimination analysis offers an effective approach to feature selection.
• The method provides a robust way to evaluate features, even with data orientation considerations.
• This technique enhances data analysis by improving feature relevance.